/**
 * HTTP entrypoint — the hosted form of the TracePass MCP server.
 *
 * Runs as a standalone Node service (deployed to Hetzner, served at
 * https://ai.tracepass.eu/mcp). An MCP client connects over
 * Streamable HTTP; the caller's TracePass credential travels in the
 * `Authorization: Bearer <token>` header of every request.
 *
 * BOTH v1 auth methods work here, transparently — we forward the
 * Bearer token to the v1 API unchanged and the platform's unified gate
 * decides: a `tp_…` API key (service account) OR an OAuth 2.0 access
 * token (user-authorized, scoped). The server neither parses nor cares
 * which; it's a pass-through. OAuth-capable MCP clients (Claude.ai,
 * ChatGPT) discover the flow via the 401 `resource_metadata` param and
 * the server card; simpler clients paste a tp_ key.
 *
 * Stateless by design: each MCP request is self-contained, so we
 * build a fresh server + transport per request, bound to that
 * request's token. No server-side session state — which means the
 * service scales horizontally and a restart drops nothing.
 *
 * This is the SAME server core (`createMcpServer`) the stdio
 * entrypoint uses — only the transport + the key source differ.
 *
 * Env:
 *   PORT                (optional) — listen port, default 8080
 *   TRACEPASS_BASE_URL  (optional) — the TracePass API base URL the
 *                       tools call, default https://app.tracepass.eu. In prod
 *                       this is the INTERNAL Docker address (platform:3000).
 *   TRACEPASS_AUTH_SERVER_URL (optional) — the PUBLIC authorization-server
 *                       origin advertised in OAuth discovery metadata, default
 *                       https://app.tracepass.eu. Set this when BASE_URL is an
 *                       internal address so clients get a reachable auth server.
 */
export {};
//# sourceMappingURL=http.d.ts.map